Density-Aware Graph for Deep Semi-Supervised Visual Recognition

Li, Suichan; Liu, Bin; Chen, Dongdong; Chu, Qi; Liu, Yuan; Yu, Nenghai

doi:10.1109/cvpr42600.2020.01341

Cited by 28 publications

(22 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…FixMatch (Sohn et al 2020) provides a simplified version where pseudo-labeling is used instead of distribution sharpening, without the need for additional tricks such as distribution alignment or augmentation anchoring (i.e., using more than one weak and one strong augmented version) from ReMixMatch or training signal annealing from UDA. Additionally, similar unlabeled images can be encouraged to have consistent pseudo-labels (Hu, Yang, and Nevatia 2021), or pseudo-labels can be propagated via a similarity graph (Li et al 2020) or centroids (Han et al 2021). Our method extends FixMatch by leveraging a self-supervised loss in cases where the pseudo-label is un-confident, allowing to perform barely-supervised learning in realistic settings.…”

Section: Semi-supervised Learningmentioning

confidence: 99%

Barely-Supervised Learning: Semi-Supervised Learning with very few labeled images

Lucas¹,

Weinzaepfel²,

Rogez³

2021

Preprint

View full text Add to dashboard Cite

This paper tackles the problem of semi-supervised learning when the set of labeled samples is limited to a small number of images per class, typically less than 10, problem that we refer to as barely-supervised learning. We analyze in depth the behavior of a state-of-the-art semi-supervised method, FixMatch, which relies on a weakly-augmented version of an image to obtain supervision signal for a more stronglyaugmented version. We show that it frequently fails in barelysupervised scenarios, due to a lack of training signal when no pseudo-label can be predicted with high confidence. We propose a method to leverage self-supervised methods that provides training signal in the absence of confident pseudolabels. We then propose two methods to refine the pseudolabel selection process which lead to further improvements. The first one relies on a per-sample history of the model predictions, akin to a voting scheme. The second iteratively updates class-dependent confidence thresholds to better explore classes that are under-represented in the pseudo-labels. Our experiments show that our approach performs significantly better on STL-10 in the barely-supervised regime, e.g. with 4 or 8 labeled images per class.

show abstract

Section: Semi-supervised Learningmentioning

confidence: 99%

Barely-Supervised Learning: Semi-Supervised Learning with very few labeled images

Lucas¹,

Weinzaepfel²,

Rogez³

2021

Preprint

View full text Add to dashboard Cite

show abstract

“…In term of pseudo labeling and leveraging the unlabeled data, we will briefly review the semi-/self-supervised methods here. The semisupervised methods include consistency regularization [46,2,43,32], entropy minimization [16,37], pseudo labeling [20,21] et al However, conventional pseudo labeling strategy works under the hypothesis that the unlabeled are of the same class space as the labeled. In another research line, the self-supervised learning attempts to learn purely on unlabeled data [31,15] or serve as an auxiliary supervision on training data [14,55].…”

Section: Related Workmentioning

confidence: 99%

Mining Latent Classes for Few-shot Segmentation

Yang¹,

Zhuo²,

Qi³

et al. 2021

Preprint

View full text Add to dashboard Cite

Few-shot segmentation (FSS) aims to segment unseen classes given only a few annotated samples. Existing methods suffer the problem of feature undermining, i.e., potential novel classes are treated as background during training phase. Our method aims to alleviate this problem and enhance the feature embedding on latent novel classes. In our work, we propose a novel joint-training framework. Based on conventional episodic training on support-query pairs, we introduce an additional mining branch that exploits latent novel classes via transferable sub-clusters, and a new rectification technique on both background and foreground categories to enforce more stable prototypes. Over and above that, our transferable sub-cluster has the ability to leverage extra unlabeled data for further feature enhancement. Extensive experiments on two FSS benchmarks demonstrate that our method outperforms previous state-ofthe-art by a large margin of 3.7% mIOU on PASCAL-5 i and 7.0% mIOU on COCO-20 i at the cost of 74% fewer parameters and 2.5x faster inference speed.

show abstract

“…In a completely different direction to network predictions, it has been shown from a classical perspective [25] that energy based models such as graphs are well suited to the task of label propagation. Therefore, several works [33], [37], [38] have shown good performance by iteratively feeding the feature representation of a neural network to a graph, performing pseudo-label generation on the graph and then using those labels to train the network. However, graphical approaches have yet to show that they can produce state-of-the-art results compared to model based approaches such as [12], [13].…”

Section: B Pseudo-labelling Techniquesmentioning

confidence: 99%

“…1) Methods which used the 13-CNN architecture [32]: Π-Model [30], Mean Teacher(MT) [32], Virtual Adversarial Training (VAT) [31], Label Propogation for Deep Semi-Supervised Learning (LP) [33], Smooth Neighbors on Teacher Graphs (SNTG) [42], Stochastic Weight Averaging(SWA) [43], Interpolation Consistency Training (ICT) [21], Dual Student [44], Transductive Semi-Supervised Deep Learning(TSSDL) [35], Density-Aware Graphs (DAG) [38] and Pseudo-Label Mixup [20]. Unfortunately, due to the natural progress in the field, each paper has different implementation choices which are not standardised.…”

Section: Evaluation Protocolmentioning

confidence: 99%

LaplaceNet: A Hybrid Energy-Neural Model for Deep Semi-Supervised Classification

Sellars,

Aviles-Rivero,

Schönlieb

2021

Preprint

View full text Add to dashboard Cite

Semi-supervised learning has received a lot of recent attention as it alleviates the need for large amounts of labelled data which can often be expensive, requires expert knowledge and be time consuming to collect. Recent developments in deep semi-supervised classification have reached unprecedented performance and the gap between supervised and semi-supervised learning is ever-decreasing. This improvement in performance has been based on the inclusion of numerous technical tricks, strong augmentation techniques and costly optimisation schemes with multi-term loss functions. We propose a new framework, LaplaceNet, for deep semi-supervised classification that has a greatly reduced model complexity. We utilise a hybrid energyneural network where graph based pseudo-labels, generated by minimising the graphical Laplacian, are used to iteratively improve a neural-network backbone. Our model outperforms state-of-the-art methods for deep semi-supervised classification, over several benchmark datasets. Furthermore, we consider the application of strong-augmentations to neural networks theoretically and justify the use of a multi-sampling approach for semi-supervised learning. We demonstrate, through rigorous experimentation, that a multi-sampling augmentation approach improves generalisation and reduces the sensitivity of the network to augmentation. Code available at https://github.com/psellcam/ LaplaceNet.

show abstract

Density-Aware Graph for Deep Semi-Supervised Visual Recognition

Cited by 28 publications

References 13 publications

Barely-Supervised Learning: Semi-Supervised Learning with very few labeled images

Barely-Supervised Learning: Semi-Supervised Learning with very few labeled images

Mining Latent Classes for Few-shot Segmentation

LaplaceNet: A Hybrid Energy-Neural Model for Deep Semi-Supervised Classification

Contact Info

Product

Resources

About